of Microprocessors

332 MUSICAL ApPLICATIONS OF MICROPROCESSORS
number generator is used to select the next note based on the specified
probabilities. The character of the music generated thus depends on the table
entries and the number of prior notes considered.
One method for filling the table is analysis of existing music. For
example, one might perform a statistical analysis ofall four note sequences in
the most popular Bach organ fugues. The data obtained could be compiled
into a table like the one just described. There would probably be numerous
combinations that did not occur in the music analyzed, so one might have to
add a "back-tracking" capability to the program. One problem with extending
the technique to consider longer sequences of notes is the tremendous
increase in table size. The analysis of most conventional music, however,
would result in a large proportion of empty (zero probability) table entries.
Thus, it may be more compact to formulate the data into a set of rules.
Besides memory savings, it is usually easier to experiment with the rules than
thousands of probability table entries.
The results of such efforts have been mildly successful in producing
interesting sequences. Pieces produced by analyzing Bach's music, for example,
may sound Bach-like for a shorr run of a few notes. However, after
listening for awhile, it becomes apparent that the music is just drifting
aimlessly and getting nowhere. Overanalysis is likely to result in whole
phrases from the analyzed material appearing in the output.
Analog Feedback Techniques
Another method of producing sequences is to use the principle of
feedback. The sequences produced, while definitely not random, are complex
and often unpredictable. The basic idea is to set up a collection of devices or
modules, each of which has an input, an output, and performs some processing
function. The modules are strung together and the output of the last
module is fed back into the input of the first. Multiple-feedback paths can
also exist. A simple sequence, even a single event, is then fed into the chain
and gets processed over and over changing some on each trip. With multiple
feedback paths, the sequence may be split and duplicated on each evolution.
One of the simplest setups is a series ofSAH modules, all driven by the
same trigger as in Fig. 10-8. A multiple-input VeA is used to selectively
mix an input from outside and one or more feedback loops. With only the
input enabled, the final output from the system is simply a delayed, sampled
version of the input. Outputs taken from intermediate states would be
identical but with differing delays. This might be useful in creating sequence
echo effects or even have a sequence playa "round" with itself.
With the end-around feedback path enabled, many possibilities exist.
One could, for example, fill the SAH chain with a short sequence of notes
(five SAHs could hold a five-note sequence), disable the input, and recircu-